xAI Launches Grok Voice Think Fast 2.0, Its Fastest Voice Model

SAN FRANCISCO — xAI has released Grok Voice Think Fast 2.0, its most capable speech-to-speech voice model yet, pairing sharper transcription and smoother conversation with a response time fast enough to feel genuinely live.

Announced on July 29, the model reasons while it speaks — listening, thinking, and talking at the same time — which lets it stay quick even as it works through complex, multi-step tasks in the background.

Faster, Sharper, and Benchmark-Leading

The headline number is speed. Grok Voice Think Fast 2.0 delivers audio in 0.70 seconds, down from 1.25 seconds on version 1.0 and well ahead of competing systems. On the Artificial Analysis Speech-to-Speech Quality Index it scored 82.9%, topping both its predecessor at 75.7% and rival models from other major labs.

Accuracy climbed too. In evaluations across thousands of short phrases in 24 languages, xAI reported a 1.5 to 2.0 times improvement in transcription accuracy over dedicated speech-to-text systems from Deepgram and ElevenLabs, with the gap widening to roughly 10 times in noisy, real-world conditions. The gains extend a rapid cadence of releases from the Grok team, which earlier this summer added 21 flagship voices and a voice agent builder.

Built for Real Conversations and Real Work

xAI trained the model to talk more like a person — shorter sentences, one question at a time, and less filler — while quietly guiding users through complex workflows. It also became far more efficient with reasoning, using roughly 60% fewer reasoning tokens than version 1.0, so tool calls often execute before the agent finishes its first sentence.

The company put the model to work on its own operations, A/B testing it on Starlink's phone support line, where it reported a meaningful lift in both sales conversion and the share of calls resolved without a human. That enterprise focus mirrors xAI's broader push to embed Grok across productivity tools, including its recent move to bring Grok to Google Workspace and Microsoft 365.

Simple Pricing and an Automatic Upgrade

xAI priced the model at $0.08 per minute of audio and kept migration painless: on August 5, the "grok-voice-latest" endpoint will automatically point to Think Fast 2.0, with existing prompts expected to work without changes. Developers who want to stay on the older version can pin it before the switch.

The release, detailed on xAI's news blog, lands as voice becomes one of the most competitive fronts in AI, with every major lab chasing natural, low-latency conversation. By combining leading benchmarks, real-world accuracy, and transparent pricing, xAI is positioning Grok Voice as a practical foundation for customer support agents, in-car assistants, and any product where people would rather talk than type.

With Grok already woven into X, Tesla vehicles, and a growing set of business tools, a faster and more reliable voice layer gives Musk's AI company one more way to put Grok in front of users — and to keep pressing its case that the fastest-moving lab in AI is the one to watch.